A Proposed Model for Mediation Analysis with Structurally Different Raters

The field of prevention relies heavily on understanding causal processes as a way of identifying potential targets for prevention and how interventions operate to achieve their effects. The use of statistical mediation analysis as a tool for prevention research is important because it helps explain how an independent variable exerts its effect on a dependent variable. Furthermore, the use of multiple methods (i.e., different sources of information) and/or multiple raters (e.g., teachers, parents, students) to assess the constructs of interest (e.g., externalizing behavior problems) in prevention science is greatly valued since multimethod studies are more informative than single method designs. With a multimethod study, it is possible to assess the extent to which the different sources of information converge and examine the specific qualities each method or rater. Thus when working with multiple raters, Eid et al. (2008) suggest distinguishing between raters with different rating perspectives (i.e., structurally different raters) of a certain behavior (e.g., child self-report, parent report, teacher report) and raters with a shared perspective (e.g., peer report) of a certain behavior in which any rater with that perspective may be used (i.e., interchangeable raters). Despite the fact that many recent studies have used multi-method measurement designs to study mediated effects, many of the approaches used to integrate multiple methods in the statistical analyses do not satisfactorily account for the use of multiple methods (i.e, structurally different versus interchangeable)  and thus have significant theoretical and empirical limitations The present research aims to address this issue by integrating modern approaches of multitrait-multimethod (MTMM) methodology with modern methods of statistical mediation analysis. A technique for analyzing a mediation model with structurally different raters will be demonstrated via its application to a real prevention dataset.